Re: ANNOUNCEMENT: New image format

John Costella schrieb:
> 12:30 pm CST, November 22, 2007
> 
> 
> I am pleased to announce the release of a new image format, which I
> have somewhat mischievously dubbed "JPEG-Clear", which is optimized
> for the following uses, among others:
> 
> * Transmission of images or image sequences compressed losslessly or
>   lossily through a limited bandwidth channel (such as the Internet),
>   where rapid, smooth, progressive, and antialiased rendering of the
>   image being downloaded is desirable (such as on a web page).
> 
> * Lossless compression of images or image sequences, particularly
>   where it is advantageous to have high-quality antialiased
>   thumbnails available rapidly, without any need for them to be
>   separately generated or stored.
> 
> * Lossy compression of images, where the ability to rapidly render
>   thumbnails or render to lower resolution is advantageous (e.g.
>   photo CDs).
> 
> * Transmission of video through a limited bandwidth channel where it
>   is desirable to dynamically subscribe to and unsubscribe from
>   additional detail according to the available bandwidth, without need
>   for replication of any information at the server.
> 
> 
> The full details are posted at:
> 
> http://assassinationscience.com/johncostella/jpegclear


First of all, I wouldn't call it JPEG unless it is coming from the JPEG.
 Strictly speaking, we don't have a trademark AFAIK, but still, I think
it is confusing and I would recommend not to do this. I don't think we
will come after you, but please, *don't do that*. Thanks!

Second point is that I think you should definitely measure compression
performance and publish test results on that. By performance I do not
only mean PSNR, but other metrics as well: PQS, M-SSIM, VDP. If you need
help on that, let me know.

Third, as you may or may not know, JPEG currently has an open call for
advanced image coding (AIC) where, if competitive, you may file your
format for discussion in the JPEG and finally can standardize it. I
highly recommend, though, to make measurements upfront.

> * Lossless compression ratios comparable to those of PNG.

You may want to compare to state of the art, which is for example
JPEG-LS. PNG is not a very reasonable lossless compressor.

> Negatives:
> ---------
> 
> * Adaptive lossy compression is up to ten times slower than lossless
>   compression or fixed lossy compression (it actually performs ten
>   times as many passes).

Have you considered implementing a smarter rate allocation, either
a posteriori as in the EBCOT in JPEG2000, or an a priori algorithm that
estimates the rate from the coefficient statistics? If it needs "passes"
to find the rate, it looks like you can do better.

> * Compression and full decompression times expected to be slower
>   than JPEG-XR (not tested).

As long as it is within the limits of JPEG2000, I wouldn't care as long
as it is reasonably better than JPEG-XR.

> * Lossy compression quality satisfactory for transmission and
>   display purposes, but not competitive with JPEG-2000 or JPEG-XR
>   for master image storage.

...which is too bad.

> * Grayscale and color, in either the RGB or YCbCr colorspace.

Please think about ICC profiles to support arbitrary color spaces.
(Besides, there is no "the RGB" color space)

So long,
	Thomas